The Research On Ultrasonic Testing Technology And Signal Processing Of Metal Sheet Defects

Metal sheet is often used in aerospace,high pressure vessels and other manufacturing components.The detection of sheet defects and the assessment of its safety performance are essential.Comparing with other conventional detection methods,the Ultrasonic Lamb wave detection technology has its advantages on effectiveness and accuracy.Thus,it has been applies in extensive use.Due to the use of the diagonal probe with a certain angle in conventional thin plate detection,the fixed angle can only form a fixed mode of ultrasound.As a result,such ultrasound can only identify the defects within a certain range,and is incapable of detecting the other major category.In order to solve the limited defects detection problem caused by the fixed mode of ultrasound,this paper has constructed a dual probe ultrasonic Lamb wave detection system based on the theory of the direct probe stimulating the multimodal ultrasonic Lamb wave detection principle.The advantage of this stimulus is that the sonic waves spread out to all directions due to the constraints of the plates.In the process of diffusion,the multimodal signals of the multimode are generated;and those signals can identify a various of defects.In the mean time,the signals are complicated,and then,needed been processed.In this paper,the author has first reviwed and analysed the multimodal detection principle of ultrasonic Lamb wave and drawn its theoretical dispersion curve by using the Rayleigh-Lamb equation which solving and describing the characteristics.When detecting the thin plate with the Lamb wave,a modal recognition is necessary.By comparing with the commonly used model recognition methods,the STFT(short-time Fourier transform)method is used in this paper;and the obtained two-dimensional time-frequency diagram is compared with the former theoretical dispersion curve so that an appropriate model can be identified and selected effectively.in addition,due to the complexity of the Lamb wave signal,it is necessary to figure out a method to reduce the signal noise effectively.Both wavelet threshold denoising and empirical mode decomposition(EMD)method are proved effective to certain degrees.By comparing the simulation results,the EMD method has been chosen in this paperfor its adaptability and better capacity of retaining signal characteristics.And then,through the analyzing of the IMF components and extracting the feature information of the time domain and frequency,the foundation for the classification of the subsequent defects has been built up.Finally,based on the foundation,two models of BP neural network and genetic algorithm optimization vector machine(GA-SVM)are established and the defect types are coded.The normalized training samples are used as input to train these two models respectively and then test their feasibility.The experimental results show that the defect recognition rate of BP neural network and GA-SVM can reach 100%.By comparing there two results comprehensively,a conclusion can be drawn that GA-SVM model is better than BP neural network model in terms of effectiveness and time-saving.